Introduction

In this notebook we train the static part of the cross-domain model. The static map in the Carla dataset is the day images. We train this by using a Siamese network and triplet loss. The non-grey parts of following image denote the siamese network as a part of the bigger model.

Static Part

Setup

Setting paths

Loading images

Creating dataloaders

Transforms

Defining dataloaders

Plotting examples

Training the Siamese Network

Training Setup

Loading the Model (Mhen Continuing Training)

Plotting PCA of Embeddings Before Training

If the interactive plot is not viewable please check train_static.html in the html folder

Training

Plotting PCA of Embeddings After Training

If the interactive plot is not viewable please check train_static.html in the html folder

It can be observed that the embeddings are better clustered in a continous manner.

Saving model

--fin--